December twenty twenty five came and went without fanfare. No breathless announcements. No revolution headlines. Just model releases from Anthropic, Google and OpenAI within six weeks of each other.
The industry barely flinched. That is how fast the baseline is moving.
This is what actually happened. Last week I handed Opus four point five a product problem I had been wrestling with for days. Market positioning for a new feature. Unclear competitive landscape. Multiple technical approaches with different tradeoffs.
The model acted as my product manager. It asked clarifying questions I had not thought to ask myself. I told it to search Google and Bing for competitive analysis. It did. It synthesized what it found and came back with a positioning strategy that accounted for gaps in competitor offerings I had not known existed.
Then it built a working prototype.
I ran the same problem through Gemini three. Different reasoning style. Equally coherent output.
Then GPT five point two. Same story.
Three frontier models from three different companies. All three understood what I was trying to accomplish well enough to contribute at every level. Market research, strategy, and implementation. Any one of these models, arriving alone, would have been a generational leap in capability.
All three arrived in a month.
Consider the gap that is opening. Walk into any engineering organization today and you will find three populations working side by side in completely different realities.
Some engineers are not using artificial intelligence at all. They are building software the same way they did in twenty twenty three.
Most are using AI occasionally. For boilerplate. For documentation. Leadership celebrates that we are using AI without recognizing the gap between using these models for autocomplete and using them to reason through complex problems. It is the difference between having electricity and having the internet.
A small but growing population has crossed a threshold. They have reorganized how they think about building software. They are shipping in weeks what used to take quarters. They are building alone what used to require teams.
The gap between these groups is not about productivity. It is about capability. And it is not a tools problem. And it is widening every month.
Here is the pivot you are missing. While everyone watched model releases, something else happened in December.
Anthropic released Agent Skills as an open standard. OpenAI adopted it. The same skill format now works across Claude, Codex and ChatGPT.
This matters because skills encode institutional knowledge. Your compliance workflows. Your architecture patterns. Your way of doing things. Once encoded, that knowledge compounds. Every process you teach the model becomes available to every person in your organization, instantly.
The companies building skills libraries today are not just getting more efficient. They are creating organizational capability that takes years to develop. The ones waiting will find themselves trying to close that gap in months.
If you are still thinking about AI as which model should we use, you are already asking the wrong question.
Twenty twenty eight is two years away. The executives hiring in twenty twenty eight will be looking for leaders who spent twenty twenty five and twenty twenty six and twenty twenty seven building with AI. Not evaluating. Not running pilots. Building.
This thing is evolving too fast for consultancies to package into adoption frameworks. By the time they publish the methodology, the capabilities have moved. This is not Agile. This is not DevOps. There is no twelve month transformation roadmap. There is only rapid evolution.
The leaders who will thrive in twenty twenty eight are the ones putting their hands on the keyboard now. Not delegating to their teams. Not waiting for the approved vendor or the security review or someone to tell them it is safe. They are building, and they are learning what these models can actually do.
December twenty twenty five was quiet. Just three model releases. Just an open standard that lets organizations encode what they know. Just a pivot point that will not be obvious until we are looking back at it.
Those who know, know.